Unknown Class Discovery for Source-Free Domain Adaptation

Abstract

Source-Free Domain Adaptation (SFDA) is a promising technology to adapt source-pretrained model to target domain without accessing source data, and most existing SFDA methods assume that the source and target domain share the same label space. Although a fraction of methods investigate how to detect unknown class samples in target domain, they usually treat all samples belonging to different unknown classes as one class. The more practical problem is not only detecting unknown classes when they exist, but also determining which samples of unknown classes belong to the same class. Due to the presence of unknown classes, we propose an adaptive cluster method to estimate the number of classes and design a two-branch network to classify known classes and discover unknown classes. Furthermore, We exploit branch consensus to select reliable samples of known and unknown classes to improve the performance. The experiments conducted on real-world datasets verify the effectiveness and superiority of our method.

Keywords

source-free domain adaptation / unknown class discovery / two-branch network / adaptive clustering / branch consensus

RIGHTS & PERMISSIONS

Higher Education Press 2026

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.